I've been frustrated that "context engineering" is everywhere but nobody specifies what it actually means. Frameworks define the tool loop but leave injection points undefined.

This is a proposal for what context engineering infrastructure should look like:

- Renderable context components (tools serving UIs vs models differently) - Queryable conversations (event stream with materialized views) - Reactive injection (rules that fire based on conversation state) - Injection queue (priority, batching, deduplication) - Hookable architecture

Blog post: https://michaellivs.com/blog/context-engineering-open-call

Repo (early design phase, looking for collaborators): https://github.com/Michaelliv/context-engine

Interested in feedback on this direction, especially: How should rules be expressed - DSL or code? What's the right integration surface for existing frameworks?